Fasting plasma glucose, diagnosis of gestational diabetes and the risk of large for gestational age: a regression discontinuity analysis of routine data.

OBJECTIVE: To estimate the causal effects of fasting plasma glucose (FPG) and diagnosis of gestational diabetes (GDM) on birthweight and the risks of large for gestational age (LGA). DESIGN: Regression discontinuity analysis of routine data. SETTING: Two district general hospitals in West Yorkshire, UK. POPULATION: A cohort of 7062 women with singleton pregnancies who were screened for GDM and gave birth to a baby at ≥24 weeks of gestation in 2017-2019, inclusive. METHODS: The causal effects of FPG and GDM diagnosis were estimated using the two-stage least-squares approach, around the diagnostic threshold of FPG ≥ 5.6 mmol/l recommended by the UK's National Institute for Health and Care Excellent (NICE), controlling for ethnicity, maternal age, parity, height and weight. MAIN OUTCOME MEASURES: Birthweight (standardised for sex and gestational age) and large for gestational age (standardised as birthweight above the 90th centile). RESULTS: For each 1 mmol/l increase in FPG the observed birthweight increased by Z-score = 0.48 standard deviations (95% CI 0.39 to 0.57) and the odds of LGA increased by OR = 2.61 (95% CI 1.86 to 3.66). Conversely, GDM diagnosis reduced the observed birthweight by Z = -0.61 (95% CI -0.94 to -0.29) and lowered the odds of LGA by OR = 0.33 (95% CI 0.15 to 0.74). Similar, but less certain, patterns were observed for caesarean section, shoulder dystocia and perinatal death. CONCLUSIONS: The relationship between FPG and LGA is potent but is dramatically reduced by GDM diagnosis (and all the consequences thereof). Women with mild hyperglycaemia (with an FPG of 5.1-5.5 mmol/l) who fall below the current NICE threshold for GDM diagnosis have the highest risks of adverse outcomes, suggesting a need to reconsider their current care. TWEETABLE ABSTRACT: Regression discontinuity analysis shows that untreated mild hyperglycaemia increases the odds of large for gestational age, but that a diagnosis of gestational #diabetes lowers the odds by three times.

Gestational diabetes and the risk of late stillbirth: a case-control study from England, UK.

OBJECTIVE: To explore the separate effects of being 'at risk' of gestational diabetes mellitus (GDM) and screening for GDM, and of raised fasting plasma glucose (FPG) and clinical diagnosis of GDM, on the risk of late stillbirth. DESIGN: Prospective case-control study. SETTING: Forty-one maternity units in the UK. POPULATION: Women who had a stillbirth ≥28 weeks of gestation (n = 291) and women with an ongoing pregnancy at the time of interview (n = 733). METHODS: Causal mediation analysis explored the joint effects of (i) 'at risk' of GDM and screening for GDM and (ii) raised FPG (≥5.6 mmol/l) and clinical diagnosis of GDM on the risks of late stillbirth. Adjusted odds ratios (aOR) were estimated by logistic regression adjusted for confounders identified by directed acyclic graphs. MAIN OUTCOME MEASURES: Screening for GDM and FPG levels RESULTS: Women 'at risk' of GDM, but not screened, experienced 44% greater risk of late stillbirth than those not 'at risk' (aOR 1.44, 95% CI 1.01-2.06). Women 'at risk' of GDM who were screened experienced no such increase (aOR 0.98, 95% CI 0.70-1.36). Women with raised FPG not diagnosed with GDM experienced four-fold greater risk of late stillbirth than women with normal FPG (aOR 4.22, 95% CI 1.04-17.02). Women with raised FPG who were diagnosed with GDM experienced no such increase (aOR 1.10, 95% CI 0.31-3.91). CONCLUSIONS: Optimal screening and diagnosis of GDM mitigate the higher risks of late stillbirth in women 'at risk' of GDM and/or with raised FPG. Failure to diagnose GDM leaves women with raised FPG exposed to avoidable risk of late stillbirth. TWEETABLE ABSTRACT: Risk of #stillbirth in gestational diabetes is mitigated by effective screening and diagnosis.

Adjustment for time-invariant and time-varying confounders in 'unexplained residuals' models for longitudinal data within a causal framework and associated challenges.

'Unexplained residuals' models have been used within lifecourse epidemiology to model an exposure measured longitudinally at several time points in relation to a distal outcome. It has been claimed that these models have several advantages, including: the ability to estimate multiple total causal effects in a single model, and additional insight into the effect on the outcome of greater-than-expected increases in the exposure compared to traditional regression methods. We evaluate these properties and prove mathematically how adjustment for confounding variables must be made within this modelling framework. Importantly, we explicitly place unexplained residual models in a causal framework using directed acyclic graphs. This allows for theoretical justification of appropriate confounder adjustment and provides a framework for extending our results to more complex scenarios than those examined in this paper. We also discuss several interpretational issues relating to unexplained residual models within a causal framework. We argue that unexplained residual models offer no additional insights compared to traditional regression methods, and, in fact, are more challenging to implement; moreover, they artificially reduce estimated standard errors. Consequently, we conclude that unexplained residual models, if used, must be implemented with great care.